Merge pull request #990 from single-cell-data/ph/fix/seurat-empty-mat…

…rices

[r] Handle empty matrices when creating SOMA from Seurat

apis/r/DESCRIPTION

@@ -7,7 +7,7 @@ Description: Interface for working with 'TileDB'-based Stack of Matrices,
     from and export to in-memory formats used by popular toolchains like
     'Seurat', 'Bioconductor', and even 'AnnData' using the companion Python
     package.
-Version: 0.1.22.9001
+Version: 0.1.22.9002
 Authors@R: c(
     person(given = "Aaron",
            family = "Wolen",

apis/r/NEWS.md

@@ -1,5 +1,9 @@
 # tiledbsoma (development version)
 
+## Changes
+
+- The `SOMA`'s `from_seurat_assay()` method now checks for and skips empty matrices following Seurat's definition of an empty matrix (#990)
+
 ## Features
 
 - The `SOMACollection`'s `to_seurat()` method gains a `somas` argument that makes it possible to select a subset of `SOMA`s and `X` layers to be retrieved (#571).

apis/r/R/SOMA.R

@@ -319,16 +319,26 @@ SOMA <- R6::R6Class(
           = inherits(object, "Assay"),
         "'var' must be a logical value" = is.logical(var)
       )
-
-      if (is.null(layers)) {
-        layers <- c("counts", "data")
-        if (seurat_assay_has_scale_data(object)) {
-          layers <- c(layers, "scale.data")
-        }
-      } else {
-        layers <- match.arg(layers, c("counts", "data", "scale.data"), TRUE)
-      }
-
+      layers <- layers %||% c('counts', 'data', 'scale.data')
+      layers <- match.arg(
+        arg = layers,
+        choices = c('counts', 'data', 'scale.data'),
+        several.ok = TRUE
+      )
+      layers <- Filter(
+        f = function(x) {
+          return(!SeuratObject::IsMatrixEmpty(x = SeuratObject::GetAssayData(
+            object = object,
+            slot = x
+          )))
+        },
+        x = layers
+      )
+      stopifnot(
+        "None of the layers specified are found in the provided assay" =
+          as.logical(x = length(x = layers))
+      )
+
       skip_obs <- self$obs$exists() && is.null(obs)
       if (!is.null(obs)) {
         stopifnot(
@@ -452,6 +462,10 @@ SOMA <- R6::R6Class(
         # CreateAssayObject only accepts a dgTMatrix matrix for `counts`, 'data'
         # and 'scale.data' must be coerced to a dgCMatrix and base::matrix,
         # respectively. Bug?
+        # PH: No, not a bug. The `counts` and `data` matrices must be either a
+        # dgCMatrix or base::matrix; we call SeuratObject::as.sparse() for
+        # `counts` and `data` if no one of the above. The `scale.data` slot
+        # must be a base::matrix and we make no attempts at conversion
         assay_obj <- SeuratObject::CreateAssayObject(
           data = as(assay_mats$data, "dgCMatrix"),
           min.cells = min_cells,

apis/r/R/utils-seurat.R

@@ -4,5 +4,6 @@ is_seurat_assay <- function(x) {
 
 seurat_assay_has_scale_data <- function(x) {
   stopifnot(is_seurat_assay(x))
-  length(SeuratObject::GetAssayData(x, "scale.data")) > 0
+  !SeuratObject::IsMatrixEmpty(SeuratObject::GetAssayData(x, 'scale.data'))
+  # length(SeuratObject::GetAssayData(x, "scale.data")) > 0
 }

apis/r/tests/testthat/test_SOMA_Seurat.R

@@ -318,3 +318,74 @@ test_that("individual layers can be added or updated", {
   expect_identical(soma$X$members$counts$fragment_count(), 1)
   expect_identical(soma$X$members$data$fragment_count(), 2)
 })
+
+test_that("assays with missing layers are handled: new('matrix')", {
+  try(tiledb::tiledb_vfs_remove_dir(tdb_uri), silent = TRUE)
+  assay <- pbmc_small[["RNA"]]
+  assay <- SeuratObject::SetAssayData(assay, "counts", new("matrix"))
+  soma <- SOMA$new(uri = tdb_uri)
+  soma$from_seurat_assay(assay)
+
+  expect_equal(names(soma$members$X$members), c("data", "scale.data"))
+  expect_s4_class(assay1 <- soma$to_seurat_assay(), "Assay")
+  expect_true(
+    inherits(mat <- SeuratObject::GetAssayData(assay1, "counts"), "matrix")
+  )
+  expect_true(all(dim(mat) == 0L))
+  expect_true(SeuratObject::IsMatrixEmpty(mat))
+})
+
+test_that("assays with missing layers are handled: new('dgCMatrix')", {
+  try(tiledb::tiledb_vfs_remove_dir(tdb_uri), silent = TRUE)
+  assay <- pbmc_small[["RNA"]]
+  assay <- SeuratObject::SetAssayData(assay, "counts", new("dgCMatrix"))
+  soma <- SOMA$new(uri = tdb_uri)
+  soma$from_seurat_assay(assay)
+
+  expect_equal(names(soma$members$X$members), c("data", "scale.data"))
+  expect_s4_class(assay1 <- soma$to_seurat_assay(), "Assay")
+  expect_true(
+    inherits(mat <- SeuratObject::GetAssayData(assay1, "counts"), "matrix")
+  )
+  expect_true(all(dim(mat) == 0L))
+  expect_true(SeuratObject::IsMatrixEmpty(mat))
+})
+
+test_that("assays with missing layers are handled: matrix()", {
+  try(tiledb::tiledb_vfs_remove_dir(tdb_uri), silent = TRUE)
+  assay <- pbmc_small[["RNA"]]
+  assay <- SeuratObject::SetAssayData(assay, "counts", matrix())
+  soma <- SOMA$new(uri = tdb_uri)
+  soma$from_seurat_assay(assay)
+
+  expect_equal(names(soma$members$X$members), c("data", "scale.data"))
+  expect_s4_class(assay1 <- soma$to_seurat_assay(), "Assay")
+  expect_true(
+    inherits(mat <- SeuratObject::GetAssayData(assay1, "counts"), "matrix")
+  )
+  expect_true(all(dim(mat) == 0L))
+  expect_true(SeuratObject::IsMatrixEmpty(mat))
+})
+
+test_that("assays with missing layers are handled: dgCMatrix()", {
+  try(tiledb::tiledb_vfs_remove_dir(tdb_uri), silent = TRUE)
+  assay <- pbmc_small[["RNA"]]
+  expect_s4_class(
+    m <- as(
+      as(Matrix::Matrix(NA_real_, sparse = TRUE), "generalMatrix"),
+      "CsparseMatrix"
+    ),
+    "dgCMatrix"
+  )
+  assay <- SeuratObject::SetAssayData(assay, "counts", new("matrix"))
+  soma <- SOMA$new(uri = tdb_uri)
+  soma$from_seurat_assay(assay)
+
+  expect_equal(names(soma$members$X$members), c("data", "scale.data"))
+  expect_s4_class(assay1 <- soma$to_seurat_assay(), "Assay")
+  expect_true(
+    inherits(mat <- SeuratObject::GetAssayData(assay1, "counts"), "matrix")
+  )
+  expect_true(all(dim(mat) == 0L))
+  expect_true(SeuratObject::IsMatrixEmpty(mat))
+})